I'm planning to convert a table from utf8 (_general_ci) to utf8mb4 (utf8mb4_unicode_ci). I want to make sure I don't have any malformed data after the conversion though so I was planning to duplicate the data, run a join on the two tables, and see if there are any differences.

Is this a good way to go about it?


Create table Email_utf8mb4  like Email;


insert into Email_utf8mb4 select * from Email;


ALTER TABLE Email_utf8mb4 
COLLATE utf8mb4_unicode_ci; 


select * from Email as old
join Email_utf8mb4 as new  
on old.notificationid =new.notificationid 
and (old.subject <> new.subject OR old.content <>

4.5. Assuming no rows are returned...


drop Email; 


RENAME TABLE Email_utf8mb4 to Email;

subject and content are the only alpha columns in my table.

1 Answer 1


That won't fix or recognize any malformed data already in the table.

Yes, that is a good technique. ALTER..CONVERT TO.. does the bulk of the work. And, since utf8 is a subset of utf8mb4, there should be no differences discovered in step 4.

However, there is still a possibility of step 4 showing something. This is because the definition of "equal" (hence <>) is different for _general_ci versus _unicode_ci.

For example, in German, ss and ß are unequal in _general_ci (either utf8 or utf8mb4), but equal in virtually all other collations. If, for example, you currently have a UNIQUE (or PRIMARY KEY) with a value that differs in just ss vs ß, the conversion will have a problem with "duplicate key" error.

Another problematic pair: ae vs æ

A different issue - not equality, but ordering - Ð < E for _unicode_ci, but not for some other collations.

Meanwhile, as long as you are changing the COLLATION, you may as well go to the newer Unicode Algorithm in utf8mb4_unicode_520_ci.

  • Malformed data already present is fine. So it sounds like step 4 might bring false positives but wouldn't miss any issues, is that right? Jul 11, 2018 at 20:19
  • @user3783243 - Give it a test run; that is, abort in 4.5 if rows were returned. Then let's discuss them.
    – Rick James
    Jul 11, 2018 at 22:38
  • No rows were returned so I think I'm good to go. Jul 12, 2018 at 14:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.